Mounir IDRASSI · mounir@amcrypto.jp
Version 1.0.1 · 22 September 2026
This release contains a public technical report and reproducible evidence about the supplied CHAMP specification. Read the PDF report or Markdown report on GitHub. For an offline web version, download REPORT.html and open it locally. The editable sources are REPORT.md and REPORT.tex.
AI-use disclosure. Moonshot AI's kimi-k3 was used for the main review. OpenAI's GPT-6 Astra was used for polishing and preparing the publication package, including additional technical analysis, cross-checking of claims, and development of reproducibility code. Mounir IDRASSI is responsible for the final report and its conclusions.
Independent preparation. The findings, derivations and experiments were completed independently of Saarinen's CHAMP report and public comments signed Tsinghua Hash Lab and ISCAS. Mounir IDRASSI became aware of these sources on 22 September 2026, after this review and its reproducibility package were complete. They are cited as related work identified afterward. See PROVENANCE.md for the chronology and FINDINGS.md for the overlap and additional contributions: the explicit subgroup search and collision certificates, demonstrated naive-MAC forgeries, related-digest and short-input recovery, and digest-length, initialization and timing findings.
The principal finding is a concrete adaptation of a subgroup collision-search strategy to CHAMP's determinant-2 matrices, with twelve reduced-parameter collision certificates. Its projected full-parameter work factors depend on mixing assumptions. The package also demonstrates forgeries against naive keyed constructions and a short-input preimage search at the full parameters, and documents specification and implementation issues. It does not demonstrate a practical full-parameter collision, a distinct second preimage, or recovery of a uniformly random long input from a single digest.
| File | Purpose |
|---|---|
| REPORT.pdf | Typeset public report, with the requested byline |
| REPORT.html | Standalone report with native MathML; no network needed for rendering |
| FINDINGS.md | Finding-to-evidence index and limits |
| REPRODUCING.md | Exact commands, dependencies and test coverage |
| PROVENANCE.md | Document versions, source locations and input hashes |
| AI_DISCLOSURE.md | AI tools used and their roles in preparing the review |
| LICENSING.md | MIT for code; CC BY 4.0 for the report and original results; scope and attribution |
| CHANGES.md | Release history and scope of this revision |
| code / data / evidence | Independent programs, complete certificates and recorded transcripts |
| ABSTRACT.txt | Plain-text abstract for a publication or repository submission |
| CITATION.cff / references.bib | Citation metadata and bibliography |
| SHA256SUMS | Checksums of the distributed files |
After extracting the archive, run:
python3 verify_package.py
python3 run.pyThe second command verifies all saved collisions and runs the quick mathematical demonstrations using only Python's standard library. REPRODUCING.md also gives the commands for the 40-bit preimage demonstration, complete seeded collision replay, C/KAT comparisons and proof-enabled SageMath parameter checks.
The original specification, implementations and KATs are not redistributed. Their exact hashes are supplied for optional conformance checks against a separately obtained copy. The mathematical model and saved certificates work without those files. The distinct author-hosted preprint is identified explicitly; the report's page references concern the supplied ten-page specification.
This package prepares material for distribution; its contents do not claim that the authors have already been contacted or that the report has undergone peer review. No public repository URL or DOI has been assigned in this package.
Copyright © 2026 Mounir IDRASSI for the original contributions. The code is licensed under the MIT License; the report, original documentation and original results under CC BY 4.0. See LICENSING.md for the file scope, attribution and third-party exceptions. The research programs are not production cryptographic software.